Ensemble learning paradigms for flow rate prediction boosting

Abstract

Abstract In developing countries, climate change has considerably affected population welfare by increasing drinking water scarcity. Global organizations and governments have initiated many drinking water supply projects to fight against this issue. Most of these projects are led by geophysical companies in partnership with drilling ventures to locate drillings expected to give the recommended flow rate (FR). Known as cheap methods, electrical resistivity profiling (ERP) and vertical electrical sounding (VES) were the most preferred. Unfortunately, the project objective was not achieved due to numerous unsuccessful drillings, thereby creating a huge loss of investments. To reduce the repercussion of unsuccessful drillings, we introduced the ensemble machine learning (EML) paradigms composed of four base learners. The aim is to predict at least 80% of correct FR in the validation set before any drilling operations. Geo-electrical features were defined from the ERP and VES and combined with the collected boreholes data to compose the binary dataset ( FR ≤ 1m3/hr and FR >1 m3/hr) for unproductive and productive boreholes respectively). Then, the dataset is transformed before feeding to the EMLs. As a result, the benchmark and the pasting EMLs performed 85% of good predictions on the validation set whereas the extreme gradient boosting and the stacking performed 86% and 87% respectively. Finally, the correct prediction of FRs will reduce the losses in investment beneficial for funders and state governments, and geophysical and drilling ventures.